Software & Systems Modeling https://doi.org/10.1007/s10270-019-00725-0 REGULAR PAPER ChronoSphere: a graph-based EMF model repository for IT landscape models Martin Haeusler1 · Thomas Trojer2 · Johannes Kessler1 · Matthias Farwick2 · Emmanuel Nowakowski1 · Ruth Breu1 Received: 1 April 2018 / Revised: 18 January 2019 / Accepted: 4 February 2019 © The Author(s) 2019 Abstract IT Landscape models are representing the real-world IT infrastructure of a company. They include hardware assets such as physical servers and storage media, as well as virtual components like clusters, virtual machines and applications. These models are a critical source of information in numerous tasks, including planning, error detection and impact analysis. The responsible stakeholders often struggle to keep such a large and densely connected model up-to-date due to its inherent size and complexity, as well as due to the lack of proper tool support. Even though modeling techniques are very suitable for this domain, existing tools do not offer the required features, scalability or flexibility. In order to solve these challenges and meet the requirements that arise from this application domain, we combine domain-driven modeling concepts with scalable graph- based repository technology and a custom language for model-level queries. We analyze in detail how we synthesized these requirements from the application domain and how they relate to the features of our repository. We discuss the architecture of our solution which comprises the entire data management stack, including transactions, queries, versioned persistence and metamodel evolution. Finally, we evaluate our approach in a case study where our open-source repository implementation is employed in a production environment in an industrial context, as well as in a comparative benchmark with an existing state-of-the-art solution. Keywords Model-driven engineering · Model repositories · Versioning · Graph database · IT landscape 1 Introduction ing engineering areas by applying models as an abstraction layer. The primary field of application for MDE has tradi- Model-driven engineering (MDE) is a discipline that aims at tionally always been software engineering [64]. However, the improving the processes, workflows and products of exist- key innovations of MDE are not domain specific. The gen- eral concept of using a metamodel to define a structure and Communicated by Dr. Ana Moreira. then instantiating it to create actual objects applies to a wide range of problems. When comparing different use cases it B Martin Haeusler becomes evident that modeling concepts tend to be employed [email protected] in areas that exhibit high complexity and heterogeneity in Thomas Trojer their domain structures, such as cloud orchestration [22], [email protected] self-adaptive software [5], automotive systems [24] or Enter- Johannes Kessler prise Architecture Management (EAM) [43]. However, there [email protected] are still many potential application areas for model-driven Matthias Farwick approaches that have barely been investigated so far. EAM [email protected] focuses exclusively on the strategic aspects of IT manage- Emmanuel Nowakowski ment. Standard metamodels (such as ArchiMate [43]) have [email protected] been developed for this domain, yet these metamodels focus Ruth Breu primarily on high-level business services and capabilities. [email protected] The actual assets (or Configuration Items (CIs) [7]) on the 1 University of Innsbruck, 6020 Innsbruck, Austria operative level are captured in a coarse-grained way that does not allow for deeper analysis, or are excluded entirely. 2 Txture GmbH, 6020 Innsbruck, Austria 123 M. Haeusler et al. Configuration items typically comprise physical servers, Landscape use case, the core implementation is domain inde- applications, databases and network infrastructure. We refer pendent and may also serve other use cases (see Sect. 9.4). to the collection of all assets in a company as the IT Land- In our inter-disciplinary efforts to realize this repository, we scape (also known as resource landscape [36]). The IT also contributed to the state-of-the-art in the database com- Landscapes of major, globally operating companies, can munity, in particular in the area of versioned data storage grow to considerable dimensions. Due to agility require- and graph versioning. We evaluate our approach in an indus- ments, they are increasingly subject to frequent evolution trial case study in collaboration with Txture GmbH.1 This and technology shifts. Recent examples include the exten- company employs our ChronoSphere implementation as the sive usage of virtualization platforms in data centers, the primary storage back-end in their commercial IT Landscape advent of cloud computing and the emergence of As-A- modeling tool. Service solutions. Furthermore, even though commonalities The remainder of this paper is structured as follows. In do exist, every company has its own architecture and vision Sect. 2, we first describe the IT Landscape use case in more behind its landscape. The terminology also varies, as there detail. We then extract the specific requirements for our solu- is no generally accepted definition across all stakeholders tion from this environment and discuss how they were derived for common terms like Service or Application. Responsible from the industrial context. Section 3 provides a high-level persons and teams often struggle in their continuous efforts overview of our approach. In Sects. 4 through 6, we discuss to properly document these landscapes due to their inherent the details of our solution. In Sect. 7, we present the appli- size and complexity. The absence of a reliable and up-to- cation of our repository in an industrial context. Section 8 date documentation can result in slow error detection, loss evaluates the performance of ChronoSphere in comparison of traceability of changes and misguided planning processes with other model repository solutions, which is followed by due to poor information situations. Ultimately, these issues a feature-based comparison of related work in several differ- can lead to problems which cause very high costs for the ent areas in Sect. 9. We conclude the paper with an outlook companies if they remain unaddressed [30,51]. to future work in Sect. 10 and a summary in Sect. 11. Sec- The need for tool support in the area of IT Landscape doc- tions 4 through 6 consist of a revised, updated and extended umentation is evident, and model engineering is well-suited version of the content presented in our previous work, mainly to provide the required concepts. However, the existing MDE [25,27,28]. The remaining sections (most notably 2, 7 and 8) tool infrastructure is insufficient when it comes to satisfying have never been published before. the requirements of this domain. Existing solutions either do not scale with the number of elements in a real-world IT Landscape documentation, do not offer the necessary anal- 2 Use case and requirement analysis ysis capabilities, or lack the flexibility needed in long-term projects. Several state-of-the-art model repositories employ The overarching goal in IT Landscape documentation is to relational databases, even though the object-relational gap is produce and maintain a model which reflects the current IT well-known to cause additional complexity and performance assets of a company and their relationships with each other. overhead. Furthermore, the required commitment to a fixed As these assets change over time, keeping this model up-to- schema across all entries impedes the ability to perform meta- date is a continuous task, rather than a one-time effort. model evolution processes without altering past revisions. In From a repository perspective, the use case of IT Land- recent years, the NoSQL family of databases has expanded, scape documentation is unique because it is both a database and graph databases in particular are an excellent fit for stor- scenario (involving large datasets) as well as a design sce- ing model data [1,4]. The central research question we focus nario where multiple users manually edit the model in a on in this paper is how to combine domain-driven model- concurrent fashion (see Fig. 1). The amount and quality ing concepts and technologies with the innovations from the of information which is available in external data sources graph database community in order to build a model reposi- depends on the degree of automation and standardization in tory which is suitable for IT Landscape documentation. the company. For companies with a lower degree of automa- In this paper, we present a solution for storing, versioning tion, users will want to edit the model manually to keep it and querying IT Landscape models called ChronoSphere. up-to-date. In companies that have a sophisticated automa- ChronoSphere is a novel open-source EMF model reposi- tion chain in place, the majority of data can be imported tory that addresses the needs of this domain, in particular into the repository without manual intervention. Typical data scalable versioning, querying and persistence. It utilizes sources involved in such a scenario are listed in Table 1. innovative database technology and is based on a modular After gathering and consolidating the required informa- architecture which allows individual elements to be used tion in a central repository, typical use cases are centered as standalone components outside the repository context. Even though ChronoSphere has been designed for the IT 1 www.txture.io. 123 ChronoSphere: a graph-based EMF model repository for IT landscape models ing the virtualization, clustering and load balancing layers in between). Root cause analysis is the inverse question: given an Application, the task is to find all Physical Servers on which the application transitively depends. This insight allows to reduce the search space in case of an incident (rang- ing from performance problems to total application outage). Finally, analyzing the history of a single element or the entire model as a whole are important use cases in IT Land- scape management.